MCP guides

Practical MCP server guides: what MCP is, setting up MCP in Claude Code, Cursor, Codex, and Claude Desktop, and how to debug and test MCP servers.

• What is MCP, and what is an MCP server?

  MCP (Model Context Protocol) is the open standard that lets AI agents like Claude, Cursor, and Codex use tools. Here's what an MCP server is and how it works.

• Claude Code MCP servers: add them, then see every tool call

  How to add MCP servers to Claude Code with claude mcp add, where the config lives, and how to record every tool call Claude Code makes so you can debug it.

• Cursor MCP: configure servers in mcp.json and see what they do

  How to add MCP servers to Cursor with mcp.json (global or per project), and how to record every MCP tool call Cursor's agent makes so you can debug it.

• Codex MCP: add servers in config.toml and record every tool call

  How to configure MCP servers for the OpenAI Codex CLI in ~/.codex/config.toml, and how to record and inspect every MCP tool call Codex makes.

• Claude Desktop MCP servers: the config file, and how to see every call

  Where claude_desktop_config.json lives on macOS and Windows, how to add MCP servers to Claude Desktop, and how to record every tool call for debugging.

• The GitHub MCP server: set it up locally and record what it does

  Run GitHub's official MCP server locally with Docker in Claude Code, Cursor, or Codex, and record every GitHub tool call your agent makes with RelayOrb.

• How to test an MCP server: record real sessions, check them in CI

  A practical way to regression-test MCP servers: record a real agent session, commit it as a fixture, and diff the live server's answers in CI with relayorb check.

• How to debug an MCP server: see every request and response

  Debug MCP servers by recording the exact JSON-RPC traffic between your agent and the server: arguments, results, tool errors, and latency for every call.